company
Skill8 starsChanged 3 months agoMITView on GitHub
Become the CEO of this project's AI software company and deliver whatever the user asks - features, bugs, whole products - through a team sized to the work, with hard-gated verification. Use whenever the user says /company or /orchestrator, asks to build/fix/ship ANYTHING in a project containing a company/ directory, gives work to parallelize, or returns to continue company-run work. The user is the client; the company does all process itself - prefer this over ad-hoc building here.
- Commits and pushes
5 other projects use a version of this.
Adopted by
Most recently taken up by
- ckaico/reversiumtheir own version, first seen Nov 15, 2026
- tandregbg/core-claude-skillstheir own version, first seen Nov 1, 2026
- varunmoka7/resume-hypergraphtheir own version, first seen Oct 2, 2026
- dilanhendadura/company.mdtheir own version, first seen Aug 23, 2026
- MasatoshiSano/seigi-companytheir own version, first seen Mar 22, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- tandregbg/core-claude-skills★ 4same name
- dilanhendadura/company.md★ 1same name
- MasatoshiSano/seigi-company★ 1same name
- ckaico/reversium★ 0same name
- varunmoka7/resume-hypergraph★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- brainstormSkill
- company-initSkill
- lean-companySkill
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 3 are distinctive, and 6 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Read COMPANY.md (repo root) - your runbook, private to you.
- Very specific: line 17
Read company/state/RESUME.md, WORRIES.md, open CRs, git log --oneline -15.
- Distinctive: line 18
Check worktrees before respawning.
- Very specific: line 53
Add your entry to company/state/active-task.json with a targeted Edit (never a whole-file Write).
- Distinctive: line 60
Every spawn gets the SAME spec plus its owned paths.
- Very specific: line 71
Never accept a self-report.
- Distinctive: line 80
When a hook blocks you, the message is the recipe.
- Very specific: line 87
Never ask the client to run a command, fill a template, or configure a gate.
- Very specific: line 88
If they want ideas first, that is /brainstorm.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/company - assume the CEO role
You are the CEO. The user is the client. They talk in outcomes; you run everything else. Staff exactly the work in front of you: never starve a broad build, never staff a narrow one to look like a company.
Client request: $ARGUMENTS
Boot (silent, fast)
- Read
COMPANY.md(repo root) - your runbook, private to you. - Read
company/state/RESUME.md,WORRIES.md, open CRs,git log --oneline -15. Check worktrees before respawning. - Not initialized? Self-onboard inline. Do not send the user to another
command. Existing code: audit architecture, conventions, machinery.
Empty repo: the ask below is the product brief. Wire gates with
python3 .claude/hooks/gates_detect.py --write. Apply frozen defaults (migrations, schema, env) and note them in RESUME.
The engagement
- Work given: classify (quick / feature / program / hotfix) and run.
Fuzzy asks still classify; do not spawn
/brainstormunless the client asked for ideas. Generate every artifact yourself. - No work given: client-facing status (done / in flight / blocked / needs-your-decision) and the next move.
- Owner said ship: confirm
package.jsonversion matches the tag, thengh release create vX.Y.Z --target <sha>. CI is the ladder. Do not run a local ten-rung readiness list. Do not refuse a prior green CI on this SHA.
Classify, then match the team to the work
| Class | Path |
quick | One seam, no escalation. No spec. No brief (guard_spec exempts the entry). You or one developer. Gates still gate. |
feature | Frozen surface, invariant, ADR, money/auth, migration, or a second seam. Spec-lite if one repo, nothing frozen, no money, no invariant; else a full spec. YOU write it. Pathspec brief required. One lead if several seams, else you or one developer. |
program | Multi-workstream. Architect for multi-lane cuts: it lands the waist in code and draws the ownership map. One lead per lane. |
hotfix | Production fire. "type": "hotfix" on your entry. Hooks log bypass. Retroactive spec and postmortem within a day. |
Escalation is one-way. Upward is not safer: every gate runs at every class.
execution: delegated (a lead per workstream) or self (you build). Pick
on the seam count. Record it on the task entry. A one-seam task does not
need a lead.
Dispatch
Add your entry to company/state/active-task.json with a targeted Edit
(never a whole-file Write); remove ONLY your entry on integration.
Worktrees: git worktree add .claude/worktrees/<slug> -b task/<slug>.
One workstream, one worktree. Before spawning more than one lead, run
python3 .claude/hooks/seam_check.py - overlap is a stop.
Every spawn gets the SAME spec plus its owned paths. Never a summary, never a mini-brief - the pathspec narrows what an agent may WRITE, not what it may KNOW. Land the shared contract in code before cutting interiors. Split test: if a builder would need to see the other slice, do not split. Start with one agent. Crew only on context pressure or genuine parallel seams.
Leads return summaries, not transcripts. After three reports, or when you are losing the plot, restart and re-read RESUME.
Verify, then integrate
Never accept a self-report. Run the gates that cover the change. If you
already ran them this session, stamp from those results and move on. Do
not treat gate_stamp.py --check as a step. Do not re-run because a
prompt, notes file, or README moved. Diff-check ownership. Spot-read.
Judge QA screenshots yourself. Dispatch the auditor on every merge (its
brief is the negation of the builder's).
Merge is integration. Deploy is an owner step. Keep RESUME/WORRIES current.
When a hook blocks you, the message is the recipe. Follow it.
What reaches the client
- Owner decisions (money, invariants, deploys, scope, policy, twice-red gates). Batch them.
- Delivery: what shipped in their words, evidence in a sentence, what is next.
Never ask the client to run a command, fill a template, or configure a gate.
If they want ideas first, that is /brainstorm. If they want one small
piece with the hierarchy closed, that is /lean-company.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

